Lexical decision with prime target working fine locally, not online

URL of experiment: https://pavlovia.org/skeptikantin/ldt_rz


I’m having a hard time with getting experiments to run online (builder, mac-version 2020.1.2, i.e., before the most recent release, Catalina [sorry!]). In the current case, I built a relatively simple lexical decision task with a prime–target sequence: a prime flashes briefly (2 frames) then a pause (10 frames) before the target is shown. The participant then has unlimited time to make a choice if the target is a word or not. The experiment is simple and contains no custom code, no fancy blocks, etc.

It works a charm locally and I am not getting any error messages. But on the server, a number of problems pop up: for the first item in both training and test phase, it shows a brief “default text”, then it sometimes shows two or more primes (in super rapid succession), and for some of the ensuing pairs it shows the wrong prime, i.e., the columns and rows in the stimuli .xlsx are mixed up (a prime of a previous pair shows up again, but not with a word it is supposed to etc.). So it does not appear to fetch the right row values. What is weird is that there is absolutely no pattern, except the “default text” at the beginning of the routine.

I suspected it may have something to do with the settings of how long the prime is shown and when the target appears (frame N vs. frames (duration)), but I’ve fiddled around with a number of settings and it always works fine locally, but does not have an effect when run on the server. I’ve even re-built the entire experiment. So I’m a little lost because it’s very difficult to narrow down the problem when it’s OK locally.

I’ve cleared cache, etc., but any hints in that direction would be super helpful. This must be something pretty simple. (Although we noticed that the time between prime and target feels different between local/online, with the latter being faster.)

If anyone has any ideas, that’d be smashing. Thanks!

Excel files (.xlsx) are not yet supported online. Use .csv, see

Hi @walter, thanks, I’ll try! I wasn’t aware, as a previous experiment ran smoothly with .xlsx files.

Hey everyone, switching to csv has had no effect, all problems persist (“default text” showing up, random pairing of prime and target, sometimes multiple primes). Any other ideas? What is even more bizarre is that the results file shows prime–target pairs that were not show to the participant.

I have deleted all pavlovia projects and moved all experiment to new folders, renamed all files, cleared cache, just to exclude the possibility that the wrong files are used. To no avail.

(The new experiment is here: https://run.pavlovia.org/skeptikantin/rz)